Issa Sugiura

LLM agents can significantly outperform passive strategies in complex economic environments, but their success hinges on active communication and negotiation.

Open-weight models lag significantly behind proprietary counterparts on a new Japanese VQA benchmark, highlighting critical gaps in chart and table understanding.

Training VLMs on Jagle, the largest Japanese multimodal dataset, not only crushes existing models on Japanese tasks, but *also* boosts English performance when combined with English data.